Re: [PRE-REVIEW PATCH 00/16] Modernize the tasklet API

[Kees Cook <keescook@chromium.org>]

On Sun, Sep 29, 2019 at 06:30:12PM +0200, Romain Perier wrote:
> Hello,
> 
> Nowadays, modern kernel subsystems that use callbacks pass the data
> structure associated with a given callback as argument to the callback.
> The data structure specific to the driver or subsystem that uses this
> callback framework is usually "derivated" from the data structure that
> is passed as argument to the callback.
> 
> The tasklet subsystem remains the one to pass callback argument as an
> arbitrary unsigned long argument (This has several issues that are
> explained in the first commit).
> 
> This series aims to improve the tasklet API and converts all the code
> that is using it. It is based on the series for timer_list at [1].
> 
> 1. https://lore.kernel.org/patchwork/patch/835464
> 
> 
> Romain Perier (16):
>   tasklet: Prepare to change tasklet callback argument type
>   crypto: ccp - Prepare to use the new tasklet API
>   mmc: renesas_sdhi: Prepare to use the new tasklet API
>   net: liquidio: Prepare to use the new tasklet API
>   chelsio: Prepare to use the new tasklet API
>   net: mvpp2: Prepare to use the new tasklet API
>   qed: Prepare to use the new tasklet API
>   isdn: Prepare to use the new tasklet API
>   scsi: pm8001: Prepare to use the new tasklet API
>   scsi: pmcraid: Prepare to use the new tasklet API
>   treewide: Globally replace tasklet_init() by tasklet_setup()
>   tasklet: Pass tasklet_struct pointer as .data in DECLARE_TASKLET
>   tasklet: Pass tasklet_struct pointer to callbacks unconditionally
>   tasklet: Remove the data argument from DECLARE_TASKLET() macros
>   tasklet: convert callbacks prototype for using struct tasklet_struct *
>     arguments
>   tasklet: Add the new initialization function permanently

This is looking really good; thank you! I think for easier review it
would make sense to break out the "special" cases (where you're changing
structures, etc) into their own patches (and not as a bulk change --
they need review by different subsystem maintainers, etc).

Then the patch phases can be:

1) Introduce new APIs and casts
2) Convert special cases include passing the tasklet as their .data
   (while also changing the prototypes and replacing tasklet_init() with
    tasklet_setup())
3) Convert DECLARE_TASKLET() users to the same
4) Manual one-off conversions of tasklet_init() -> tasklet_setup()
5) Mechanical mass conversion of tasklet_init() -> tasklet_setup()
6) Mass removal of .data argument from DECLARE_TASKLET()
7) tasklet API internal swap and removal of .data
8) tasklet_init() and helper cast removals.

Step 1 needs to happen in an -rc1 (e.g. v5.5-rc1).

Then steps 2, 3, and 4 can happen simultaneously across all the
maintainers that need to be aware of it and land in the next release
(the linux-next for v5.6).

Finally steps 5, 6, 7, and 8 happen in the next release's -rc1
(v5.6-rc1).

If we can get the "phase 1" patch ready quick, maybe we can get into
-rc2 for v5.4 and move things up by a release...

-Kees

-- 
Kees Cook